kpuatamazon commented on issue #18699: URL: https://github.com/apache/incubator-mxnet/issues/18699#issuecomment-664253978
Ran these in a loop over the weekend with 136 samples. keep 83b5170 | a | b | c | justc++ | -- | -- | -- | -- | -- | -- 301.324 | 298.468 | 294.094 | 301.243 | 289.476 | AVG 19.978 | 20.590 | 19.759 | 20.746 | 19.351 | Sample stddev 266.168 | 262.355 | 260.640 | 266.384 | 255.469 | Min 287.478 | 284.945 | 279.012 | 288.021 | 278.632 | Q1 307.157 | 303.188 | 297.966 | 306.039 | 294.594 | Median Running a 1-tailed t-test assuming the same variance, we can say that the new version of `python/mxnet/gluon/parameter.py` (included in "keep 83b5170") is slower than the old version (in "b") with 99.8% confidence. There may also be other effects like a) seems to be slower than justc++ but I can't say it's significantly slower. [Raw data](https://github.com/apache/incubator-mxnet/files/4981297/timing.txt) ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org
